Process Monitoring and Fault Detection with Improved Singular Spectrum Analysis

Abstract

Singular Spectrum Analysis (SSA) is a dataadaptive multimodal tool for monitoring and detecting faults in chemical processes. However, it has been shown in previous studies that decomposition of process signals using SSA resulted in a larger number of low and high frequency modes to monitor the signal for the detection of faults. Although process monitoring methods based on SSA are able to detect faults effectively, in some cases the selection of an increased number of modes prior to the application of principal component analysis is computationally complex. Moreover, the increased computational complexity due to the larger number of high frequency modes due to residual noises delays the process of fault detection. Therefore a modified method for SSA signal reconstruction is proposed in the study to enhance the capability of standard SSA in process monitoring. Application on the benchmark Tennessee Eastman process showed that improved SSA approach provided better overall performance than basic SSA. The proposed algorithm is applied to three faults that were not detected effectively by the basic SSA method in the previous studies.